At each z0 the weighted least-squares fit of y on (1, X) with weights
K(|z - z0| / lambda) gives alpha(z0), beta(z0) (ESL eqs 6.16-6.17).
Usage
morie_esl_varying_coef(X, z, y, z0, lambda, kernel = "epanechnikov")
Arguments
- X
Predictors, n by q.
- z
Variable the coefficients vary with.
- y
Response.
- z0
Targets.
- lambda
Bandwidth.
- kernel
"epanechnikov", "tri-cube" or "gaussian".
Value
Named list: coefficients (one row per target), n_in_window.
References
Hastie, Tibshirani & Friedman (2009), sec. 6.4.2.
Examples
i <- 1:40
X <- cbind(cos(2 * i))
z <- ((7 * i) %% 40) / 40
morie_esl_varying_coef(X, z, 1 + z * X[, 1], c(0.3, 0.7), 0.35)$coefficients
#> [,1] [,2]
#> [1,] 1.0014045 0.3340463
#> [2,] 0.9870295 0.7236923
